Explicitly Minimizing the Blur Error of Variational Autoencoders
This work addresses blurry outputs in VAEs, a common issue in generative modeling, but is incremental as it modifies an existing loss function.
The paper tackles the problem of blurry generated samples and reconstructions in variational autoencoders by proposing a new reconstruction term that penalizes blur while still maximizing the evidence lower bound. It demonstrates improved performance over recent methods on three datasets.
Variational autoencoders (VAEs) are powerful generative modelling methods, however they suffer from blurry generated samples and reconstructions compared to the images they have been trained on. Significant research effort has been spent to increase the generative capabilities by creating more flexible models but often flexibility comes at the cost of higher complexity and computational cost. Several works have focused on altering the reconstruction term of the evidence lower bound (ELBO), however, often at the expense of losing the mathematical link to maximizing the likelihood of the samples under the modeled distribution. Here we propose a new formulation of the reconstruction term for the VAE that specifically penalizes the generation of blurry images while at the same time still maximizing the ELBO under the modeled distribution. We show the potential of the proposed loss on three different data sets, where it outperforms several recently proposed reconstruction losses for VAEs.